A Computational Usage-Based Model for Learning General Properties of Semantic Roles

We present a Bayesian model of early verb learning that acquires a general conception of the semantic roles of predicates based only on exposure to individual verb usages. The model forms probabilistic associations between the semantic properties of arguments, their syntactic positions, and the semantic primitives of verbs. Because of the model’s Bayesian formulation, the roles naturally shift from verb-specific to highly general properties. The acquired role properties are a good intuitive match to various roles, and are useful in guiding comprehension in the face of ambiguity. Learning and Use of Semantic Roles Semantic roles, such as Agent, Theme, and Recipient in (1) and (2) below, are a critical aspect of linguistic knowledge because they indicate the relations of the participants in an event to the main predicate. (1) MomAgent gave thisTheme to herRecipient. (2) MomAgent gave herRecipient thisTheme. Moreover, it is known that people use the associations between roles and their syntactic positions to help guide on-line interpretation (e.g., Trueswell et al., 1994). How children acquire this kind of complex relational knowledge, which links predicate-argument structure to syntactic expression, is still not well understood. Fundamental questions remain concerning how semantic roles are learned, and how associations are established between roles and the grammatical positions the role-bearing arguments appear in. Early theories suggested that roles are drawn from a pre-defined inventory of semantic symbols or relations, and that innate “linking rules” that map roles to sentence structure enable children to infer associations between role properties and syntactic positions (e.g., Pinker, 1989). However, numerous questions have been raised concerning the plausibility of innate linking rules for language acquisition (e.g., Fisher, 1996; Kako, 2006). An alternative, usage-based view is that children acquire roles gradually from the input they receive, by generalizing over individually learned verb usages (e.g., Lieven et al., 1997; Tomasello, 2000). For instance, Tomasello (2000) claims that, initially, there are no general labels such as Agent and Theme, but rather verbspecific concepts such as ‘hitter’ and ‘hittee,’ or ‘sitter’ and ‘thing sat upon.’ Recent experimental evidence Such elements are also termed participant, thematic, or case roles, and more or less fine-grained semantic distinctions are attributed to them. We use the widely accepted labels such as Agent and Theme for ease of exposition. confirms that access to general notions like Agent and Theme is age-dependent (Shayan, 2006). It remains unexplained, though, precisely how verb-specific roles metamorphose to general semantic roles. Moreover, experiments with children have revealed the use of verb-specific biases in argument interpretation (Nation et al., 2003), as well as of strong associations between general roles and syntactic positions (e.g., Fisher, 1996, and related work). However, specific computational models of such processes have been lacking. We have proposed a usage-based computational model of early verb learning that uses Bayesian clustering and prediction to model language acquisition and use. Our previous experiments demonstrated that the model learns basic syntactic constructions such as the transitive and intransitive, and exhibits patterns of errors and recovery in their use, similar to those of children (Alishahi and Stevenson, 2005a,b). A shortcoming of the model was that roles were explicit labels, such as Agent, which were assumed to be “perceptible” to the child from the scene. In this paper, we have extended our model to directly address the learning and use of semantic roles. Our Bayesian model associates each argument of a predicate with a probability distribution over a set of semantic properties—a semantic profile. We show that initially the semantic profiles of an argument position yield verb-specific conceptualizations of the role associated with that position. As the model is exposed to more input, these verb-based roles gradually transform into more abstract representations that reflect the general properties of arguments across the observed verbs. The semantic profiles that we use are drawn from a standard lexical resource (WordNet; Miller, 1990), so that our results are not biased toward any theory of semantic roles. One limitation of this approach is that the profiles fail to reflect any event-specific properties that an argument might have. Such properties (like “causally affected”) are almost certainly required in an accurate representation of roles, as in Dowty (1991). Despite their absence, we are able to show that intuitive profiles can be learned for each role from examples of its use. We further establish that such representations can be useful in guiding the argument interpretation of ambiguous input, an ability experimentally demonstrated in children in recent work (Nation et al., 2003). Related Computational Work A number of computational approaches for learning the selectional preferences of a verb first initialize WordNet concepts with their frequency of use as the particular argument of a verb, and then find the appropriate level in the WordNet hierarchy for capturing the verb’s restrictions on that argument (e.g., Resnik, 1996; Clark and Weir, 2002). However, none of these models generalize their acquired verb-based knowledge to a higher level, yielding constraints on the arguments of general constructions such as the transitive or intransitive. Many computational systems model human learning of the assignment of general roles to sentence constituents, using a multi-feature representation of the semantic properties of arguments (e.g., McClelland and Kawamoto, 1986; Allen, 1997). Others learn only verb-specific roles that are not generalized (e.g., Chang, 2004). As in our earlier work, these models require explicit labelling of the arguments that receive the same role in order to learn the association of the roles to semantic properties and/or syntactic positions. In the work presented here, we show that our extended model can learn general semantic profiles of arguments, without the need for role-annotated training data.